Ship agents that do the work
A multi-agent SDK, runtime, and tool harness with OpenAI compatibility. Build, orchestrate, and run autonomous agents that call tools, use memory, and coordinate — in Python or TypeScript.
Open source (Apache-2.0). Run it locally, self-host, or deploy on Hanzo Cloud.
One SDK for the whole agent loop
The pieces you would otherwise stitch together — model calls, tools, memory, and coordination — as one coherent, debuggable system.
SDK
Define agents, tools, and handoffs in a few lines — a typed API in Python and TypeScript that reads like the mental model.
Runtime
A durable runtime that executes tool calls, retries, and multi-step plans — with tracing you can actually debug.
Tool harness
Give agents real capabilities — MCP tools, code execution, browser, and your own functions — behind one safe, uniform interface.
From prototype to production agents
OpenAI-compatible
Point the SDK at any OpenAI-compatible endpoint — Enso, Zen, or frontier models — with no rewrites.
Multi-agent orchestration
Coordinate specialists with handoffs, roles, and shared memory instead of one overloaded prompt.
MCP-native tools
Every Model Context Protocol tool is available to your agents out of the box.
Memory
Short- and long-term memory backed by Hanzo Vector for retrieval-grounded agents.
Tracing & evals
See every step, token, and tool call; evaluate runs to catch regressions before users do.
Guardrails
Scope tools, sandbox code, and gate actions so autonomy stays inside the lines.
Build your first agent today
Start from the docs, or deploy an agent on Hanzo Cloud in minutes.